The objective of this project is to take the digitized Prokudin-Gorskii plates and align the RGB plates to reproduce colored images.
The single scale implementation of the RGB alignment utilized an exhaustive search (a sliding window?) over a fixed 15 by 15 pixel box. In this approach, the box will compare every single possibility of the shifting blue and red channels and iteratively trying to find a better matching with the green channel. Here I used Sum of Squared Difference as the comparison metrix and by the end, this metric should be minimized. This approach works well for smaller sized images but consumes much more time for larger tif formatted images.
The multi scale implementation first crops the unnecessary boarders and align the image pyramids. By iteratively calling the single scale algorithm, the pyramids become more aligned. Each time one layer of the image pyramid is scaled down by a factor of 2. I did not use NCC because I found SSD working perfectly fine in this case. The algorithm holds the green channel constant and aligns blue and red, resulting in faster alignment.
cathedral r [7 1] b [-5 -2]
emir r [58 18] b [-49 -24]
harvesters r [64 -3] b [-59 -17]
icon r [49 5] b [-41 -18]
lady r [60 5] b [-52 -8]
monastery r [6 1] b [ 3 -2]
nativity r [ 5 -1] b [-3 -1]
self_portrait r [93 5] b [-78 -29]
settlers r [ 8 -1] b [-7 0]
three_generations r [60 -4] b [-48 -15]
train r [43 26] b [-42 -6]
turkmen r [61 7] b [-57 -22]
village r [73 10] b [-64 -13]
Using Single Scale Algorithm
Using Multi Scale Algorithm
